This study aimed to establish a feasible model for analysing factors affecting piglet litter performance at birth. Data of 61,984 litters were collected from 16 herds, and general linear model (GLM), multilevel Poisson regression model (MPM) and multilevel linear model (MLM) were established to compare their goodness of fit for these data. Influencing factors of piglet litter performance at birth were analysed using the established optimal model. Results showed the intraclass correlation coefficients of total born piglets (TBP), piglets born alive (PBA), low‐birth‐weight piglets (LBW), and average birth weight of piglets (ABW) reached 27.89%, 23.88%, 24.66% and 22.27%, respectively (p < .05). Akaike's information criterion and Bayesian information criterion in MLM of TBP, PBA, LBW and ABW were lower than those in GLM. Pearson residuals in MPM increased to nearly 1 after introduction of a discrete scale factor, and the p values in MPM were similar to those in MLM. Analyses of MLM indicated crossbred sows with good management supplemented with oregano essential oil and farrowing at warm season had higher TBA, PBA and ABW, but lower LBW than other sows (p < .05). In conclusion, MLM is superior to GLM and can replace MPM in analysing discrete data with hierarchical structure in pig production. More importantly, other potential influencing factors of litter performance at birth can be analysed using the established MLM in the future.

中文翻译：

---

建立多层次线性模型以分析影响仔猪出生时产仔性能的因素

这项研究旨在建立一个可行的模型来分析影响仔猪出生时产仔性能的因素。从16个牛群中收集了61984窝垫料的数据，并建立了一般线性模型（GLM），多级泊松回归模型（MPM）和多级线性模型（MLM），以比较它们对这些数据的适应性。使用建立的最佳模型分析了仔猪出生时产仔性能的影响因素。结果显示，总出生仔猪（TBP），活产仔猪（PBA），低出生体重仔猪（LBW）和仔猪平均出生体重（ABW）的类内相关系数分别达到27.89％，23.88％，24.66％和分别为22.27％（p <.05）。TBP，PBA，LBW和ABW的MLM中的Akaike信息准则和贝叶斯信息准则均低于GLM中的Akaike信息准则和贝叶斯信息准则。引入离散比例因子后，MPM中的Pearson残差增加到接近1 ，MPM中的p值类似于MLM中的p值。对传销的分析表明，交配良好管理的杂种母猪补充牛至精油，在温暖季节产仔的TBA，PBA和ABW较高，但LBW较其他母猪低（p  <.05）。总之，MLM优于GLM，在生猪生产中具有层次结构的离散数据分析中可以替代MPM。更重要的是，将来可以使用已建立的传销来分析出生时产仔性能的其他潜在影响因素。